{
  "scope": "Same twelve curated states, review instructions and criteria as the local Factorio recording, serialized in one hosted prompt. Models do not control the game. Local inputs were developed on this demo; this is not held-out accuracy. Only a label is requested. Low reasoning effort, 4096 combined completion-token cap; reasoning tokens remain billed where used. One call at a time, models interleaved within each case. API time includes network and full response. Local warm times are reused from the recorded run. No model load, adapter, game, training or operating cost is included. Provider hardware is undisclosed. Failures are retained.",
  "local_cost_assumptions": {
    "hourly_usd": 0.5,
    "utilization": 1.0
  },
  "local": {
    "name": "OpenAuditor 0.6B",
    "reviews": 12,
    "policy_matches": 12,
    "api_calls": 0,
    "paid_api_tokens": 0,
    "api_charge_usd": 0,
    "input_tokens": 833,
    "candidate_token_evaluations": 3746,
    "output_tokens": 0,
    "total_seconds": 1.6389848340768367,
    "median_seconds": 0.156001812487375,
    "estimated_cost_usd": 0.00022763678251067177,
    "estimated_cost_per_1000": 0.018969731875889313
  },
  "hosted": [
    {
      "model": "anthropic/claude-fable-5",
      "name": "Fable 5",
      "reviews": 12,
      "policy_matches": 12,
      "errors": 0,
      "input_tokens": 2637,
      "output_tokens": 43,
      "reasoning_tokens": 0,
      "cost_usd": 0.02852,
      "cost_per_1000": 2.3766666666666665,
      "median_seconds": 3.0066796459723264,
      "total_seconds": 36.2628560430021,
      "paid_api_tokens": 2680,
      "cost_ratio_to_local": 125.2873093945745,
      "median_speed_ratio": 19.273363546437338
    },
    {
      "model": "openai/gpt-6-astra",
      "name": "GPT-6 Astra",
      "reviews": 12,
      "policy_matches": 12,
      "errors": 0,
      "input_tokens": 1675,
      "output_tokens": 75,
      "reasoning_tokens": 13,
      "cost_usd": 0.0205,
      "cost_per_1000": 1.7083333333333333,
      "median_seconds": 1.3415259580069687,
      "total_seconds": 22.774335789959878,
      "paid_api_tokens": 1750,
      "cost_ratio_to_local": 90.05574483130356,
      "median_speed_ratio": 8.599425459339049
    },
    {
      "model": "deepseek/deepseek-v4-pro-0813",
      "name": "DeepSeek V4 Pro",
      "reviews": 12,
      "policy_matches": 12,
      "errors": 0,
      "input_tokens": 1620,
      "output_tokens": 901,
      "reasoning_tokens": 877,
      "cost_usd": 0.0025107984,
      "cost_per_1000": 0.2092332,
      "median_seconds": 1.8399821250059176,
      "total_seconds": 23.7119012089679,
      "paid_api_tokens": 2521,
      "cost_ratio_to_local": 11.029844879670499,
      "median_speed_ratio": 11.794620175677926
    }
  ]
}